Design and statistical analysis of observational studies


Authors’ Reply


We are writing in reply to the comments raised by Dr Prashanth on our recently published paper.[1] Our study design did ‘match’ participants prospectively. And, as such, we tend to agree that our study is best thought of as a ‘prospectively matched case–control study’. Dr Prashanth is primarily making this point to lead to his next point, that the wrong statistical tests (unpaired) were used and reported in the paper because this was a matched analysis with dependent groups. As far as his second point, this is a topic of significant debate and there are two schools of thought on this. This is particularly true of the propensity matching literature where the matched cohorts are usually identified retrospectively using modelling methods. In our study, we did that prospectively. But in either case, the claim is the same that since we have matched cohorts then the sets are dependent and paired statistical tests are needed.[2] There are statisticians on either end of the debate, however. Some, in fact, believe that paired tests should not be used and unpaired tests are best used instead.[3-5]

Re-analysis of our study using paired tests to compare and discuss the potential findings with what has already been reported is an intriguing and probably worthwhile prospect, but is surely well outside the scope of this letter.